... and thus retire debug.kdb.stop_cpus tunable/sysctl.
The knob was to work around CPU stopping issues, which since have been
either fixed or greatly reduced. kdb should really operate in a special
environment with scheduler stopped and interrupts disabled to provide
deterministic debugging.
Discussed with: attilio, rwatson
X-MFC after: 2 months or never
... and also increase the timeout.
It's better to try to proceed somehow despite stuck CPUs than to hang
indefinitely. Especially so during shutdown and when entering kdb or panic.
Timeout value is still an aribitrary value.
Timeout diagnostic is just a printf; the work on something more
debuggable is planned by attilio. Need to be careful here as
stop_cpus_hard is called very early while enetering kdb and soon(-ish)
it may become called very early when entering panic.
Reviewed by: attilio
MFC after: 2 months
processors unless the invariant TSC bit of CPUID is set. Intel processors
may stop incrementing TSC when DPSLP# pin is asserted, according to Intel
processor manuals, i. e., TSC timecounter is useless if the processor can
enter deep sleep state (C3/C4). This problem was accidentally uncovered by
r222869, which increased timecounter quality of P-state invariant TSC, e.g.,
for Core2 Duo T5870 (Family 6, Model f) and Atom N270 (Family 6, Model 1c).
Reported by: Fabian Keil (freebsd-listen at fabiankeil dot de)
Ian FREISLICH (ianf at clue dot co dot za)
Tested by: Fabian Keil (freebsd-listen at fabiankeil dot de)
- Core2 Duo T5870 (C3 state available/enabled)
jkim - Xeon X5150 (C3 state unavailable)
Modify the "alternate break sequence" detecting state
machine so that only a contiguous invocation of the
break sequence is accepted. The old implementation
did not reset the state machine when detecting an
unexpected character.
While here, use an enum for the states of the machine
instead of magic numbers.bmitted by:
Sponsored by: Spectra Logic Corporation
sys/kern/kern_conf.c:
Add make_dev_physpath_alias(). This interface takes
the parent cdev of the alias, an old alias cdev (if any)
to replace with the newly created alias, and the physical
path string. The alias is visiable as a symlink to the
parent, with the same name as the parent, rooted at
physpath in devfs.
Note: make_dev_physpath_alias() has hard coded knowledge of the
Solaris style prefix convention for physical path data,
"id1,". In the future, I expect the convention to change
to allow "physical path quality" to be reported in the
prefix. For example, a physical path based on NewBus
topology would be of "lower quality" than a physical path
reported by a device enclosure.
Sponsored by: Spectra Logic Corporation
device node has been created, pass MAKEDEV_CHECKNAME in so that the devfs
code will do the check.
Use a regular static variable as before, that's good enough to keep us from
calling into devfs most of the time.
Suggested by: kib
MFC after: 1 week
Sponsored by: Spectra Logic Corporation
In devstat_new_entry(), there is no need to initialize the queue
and the mutex in this function. There are ways to do static
initialization on both, so use STAILQ_HEAD_INITIALIZER and
MTX_SYSINIT to initialize the queue and the mutex.
In devstat_alloc(), use an atomic test and set routine to guard
making our entry in /dev. Using just a plain static variable
creates a race condition on multiprocessor machines. If you
attempt to create a second entry in devfs, the kernel will panic.
Submitted by: kdm
Reviewed by: gibbs
Sponsored by: Spectra Logic Corporation
MFC after: 1 week.
interleaving.
Signal dumping to happen only for the first panic which should be the
most important.
Sponsored by: Sandvine Incorporated
Submitted by: Nima Misaghian (nmisaghian AT sandvine DOT com)
MFC after: 2 weeks
Otherwise, p_bufr is set to garbage on the stack, and if that garbage
happens to be non-NULL, and the TOLOG or TOCONS flag is set, putbuf()
will get called and attempt to fill the non-existent buffer.
This is really only relevant for tprintf() (and only when the priority is
not -1), but set it in uprintf() and ttyprintf() for completeness.
The next step, to avoid log buffer scrambling, would be to add the
PRINTF_BUFR_SIZE code to tprintf(), but this should prevent panics.
Submitted by: rmacklem
Found by: pho
for it. Do not not expect a developer to call doadump(). Calling
doadump does not necessarily work when it's declared static. Nor
does it necessarily do what was intended in the context of text
dumps. The dump command always creates a core dump.
Move printing of error messages from doadump to the dump command,
now that we don't have to worry about being called from DDB.
In msgbuf_reinit() and msgbuf_init(), we weren't initializing the mutex.
Depending on the contents of memory, the LO_INITIALIZED flag might be
set on the mutex (either due to a warm reboot, and the message buffer
remaining in place, or due to garbage in memory) and in that case, with
INVARIANTS turned on, we would trigger an assertion that the mutex had
already been initialized.
Fix this by bzeroing the message buffer mutex for the _init() and _reinit()
paths.
Reported by: mdf
This introduce all the underlying support for making this possible (via
the function cpusetobj_strscan() and keeps ktr_cpumask exported. sparc64
implements its own assembly primitives for tracing events and needs to
properly check it. Anyway the sparc64 logic is not implemented yet due
to lack of knowledge (by me) and time (by marius), but it is just a
matter of using ktr_cpumask when possible.
Tested and fixed by: pluknet
Reviewed by: marius
While we have had a fix in place (options PRINTF_BUFR_SIZE=128) to fix
scrambled console output, the message buffer and syslog were still getting
log messages one character at a time. While all of the characters still
made it into the log (courtesy of atomic operations), they were often
interleaved when there were multiple threads writing to the buffer at the
same time.
This fixes message buffer accesses to use buffering logic as well, so that
strings that are less than PRINTF_BUFR_SIZE will be put into the message
buffer atomically. So now dmesg output should look the same as console
output.
subr_msgbuf.c: Convert most message buffer calls to use a new spin
lock instead of atomic variables in some places.
Add a new routine, msgbuf_addstr(), that adds a
NUL-terminated string to a message buffer. This
takes a priority argument, which allows us to
eliminate some races (at least in the the string
at a time case) that are present in the
implementation of msglogchar(). (dangling and
lastpri are static variables, and are subject to
races when multiple callers are present.)
msgbuf_addstr() also allows the caller to request
that carriage returns be stripped out of the
string. This matches the behavior of msglogchar(),
but in testing so far it doesn't appear that any
newlines are being stripped out. So the carriage
return removal functionality may be a candidate
for removal later on if further analysis shows
that it isn't necessary.
subr_prf.c: Add a new msglogstr() routine that calls
msgbuf_logstr().
Rename putcons() to putbuf(). This now handles
buffered output to the message log as well as
the console. Also, remove the logic in putcons()
(now putbuf()) that added a carriage return before
a newline. The console path was the only path that
needed it, and cnputc() (called by cnputs())
already adds a carriage return. So this
duplication resulted in kernel-generated console
output lines ending in '\r''\r''\n'.
Refactor putchar() to handle the new buffering
scheme.
Add buffering to log().
Change log_console() to use msglogstr() instead of
msglogchar(). Don't add extra newlines by default
in log_console(). Hide that behavior behind a
tunable/sysctl (kern.log_console_add_linefeed) for
those who would like the old behavior. The old
behavior led to the insertion of extra newlines
for log output for programs that print out a
string, and then a trailing newline on a separate
write. (This is visible with dmesg -a.)
msgbuf.h: Add a prototype for msgbuf_addstr().
Add three new fields to struct msgbuf, msg_needsnl,
msg_lastpri and msg_lock. The first two are needed
for log message functionality previously handled
by msglogchar(). (Which is still active if
buffering isn't enabled.)
Include sys/lock.h and sys/mutex.h for the new
mutex.
Reviewed by: gibbs
be brought up in the order they are enumerated in the device tree (in
particular, that thread 0 on each core be brought up first). The SLIST
through which we loop to start the CPUs has all of its entries added with
SLIST_INSERT_HEAD(), which means it is in reverse order of enumeration
and so AP startup would always fail in such situations (causing a machine
check or RTAS failure). Fix this by changing the SLIST into an STAILQ,
and inserting new CPUs at the end.
Reviewed by: jhb
larger than the receive buffer, we have to receive in sections.
When notifying the protocol that some data has been drained the
lock is released for a moment. Returning we block waiting for the
rest of data. There is a race, when data could arrive while the
lock was released and then the connection stalls in sbwait.
Fix this by checking for data before blocking and skip blocking
if there are some.
PR: kern/154504
Reported by: Andrey Simonenko <simon@comsys.ntu-kpi.kiev.ua>
Tested by: Andrey Simonenko <simon@comsys.ntu-kpi.kiev.ua>
Reviewed by: rwatson
Approved by: kib (co-mentor)
MFC after: 2 weeks
Specifically, a critical_exit() call that drops the nesting level to zero
has a brief window where the pending preemption flag is set and the
nesting level is set to zero. This is done purposefully to avoid races
where a preemption scheduled by an interrupt could be lost otherwise (see
revision 144777). However, this does mean that if an interrupt fires
during this window and enters and exits a critical section, it may preempt
from the interrupt context. This is generally fine as the interrupt code
is careful to arrange critical sections so that they are not exited until
it is safe to preempt (e.g. interrupts EOI'd and masked if necessary).
However, the SMP rendezvous IPI handler does not quite follow this rule,
and in general a rendezvous can never be preempted. Rendezvous handlers
are also not permitted to schedule threads to execute, so they will not
typically trigger preemptions. SMP rendezvous handlers may use
spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks,
but using a spinlock also enters and exits a critical section. If the
interrupted top-half code is in the brief window of critical_exit() where
the nesting level is zero but a preemption is pending, then releasing the
spinlock can trigger a preemption. Because we know that SMP rendezvous
handlers can never schedule a thread, we know that a critical_exit() in
an SMP rendezvous handler will only preempt in this edge case. We also
know that the top-half thread will happily handle the deferred preemption
once the SMP rendezvous has completed, so the preemption will not be lost.
This makes it safe to employ a workaround where we use a nested critical
section in the SMP rendezvous code itself around rendezvous action
routines to prevent any preemptions during an SMP rendezvous. The
workaround intentionally avoids checking for a deferred preemption
when leaving the critical section on the assumption that if there is a
pending preemption it will be handled by the interrupted top-half code.
Submitted by: mlaier (variation specific to rm_cleanIPI())
Obtained from: Isilon
MFC after: 1 week
now the preferred typical return value from a probe routine. Discourage
the use of 0 (BUS_PROBE_SPECIFIC) as it should be used very rarely.
Point the reader to the DEVICE_PROBE(9) manpage for more detailed notes
on possible probe return values.
Submitted by: Philip Soeberg philip-dev of soeberg net
least significant cpuset_t word at the outmost right part of the string
(more far from the beginning of it). This follows the natural build of
bits rappresentation in the words.
method, so that callers can indicate the minimum vnode
locking requirement. This will allow some file systems to choose
to return a LK_SHARED locked vnode when LK_SHARED is specified
for the flags argument. This patch only adds the flag. It
does not change any file system to use it and all callers
specify LK_EXCLUSIVE, so file system semantics are not changed.
Reviewed by: kib
